It’s in Your Genes: How Genes Explain Alcohol Consumption and Labor-Market Outcomes∗ Daiji Kawaguchi†

Jungmin Lee‡

Izumi Yokoyama§

March 25, 2018

Abstract Drinking is considered to be part of corporate culture in Japan and Korea, but about one half of the population is genetically incapable of digesting alcohol because of a deficiency of aldehyde dehydrogenase (ALDH), which dissolves toxic acetaldehyde resulting from alcohol intake. We conducted surveys in Japan and Korea that cover about 2,000 and 500 prime-age men in each country, respectively; these surveys requested biomarker information on alcohol tolerance, along with the frequency and amount of alcohol consumption, wages and information on other demographic variables. We find that the genetic disposition of alcohol tolerance increases both alcohol consumption and hourly wages. The findings, however, do not indicate a causal impact of alcohol intake on wages. We find suggestive evidence that the exclusion restriction for the Mendelian randomization is violated due to the heterogeneity in the treatment effects depending on the genetic ability. JEL Classification: C41; D12; I19 Keywords: Drinking; Wages; Genotype; Alcohol Patch Test

∗

This project has been financially supported by the Japan Science and Technology Agency (JST) and the Japan Society of Promotion of Science (JSPS), grant number 16K13369. Lee’s work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2017S1A3A2066494). The Japanese and Korean web surveys in this study were conducted by Intage and CNR Korea, respectively. Dr. Yokoi of Lifecare Giken Corporation provided the alcohol patch used for the Korean survey and technical advice. We appreciate the help of all those who have supported this project. We appreciate comments from Jerome Adda, John Cawley, Yingying Dong, Hitoshi Shigeoka, Daniel Hamermesh, Keisuke Hirano, Yoko Ibuka, Hidehiko Ichimura, Thoshiaki Iizuka, Hyuncheol Kim, Chulhee Lee, Yoshito Takasaki, and seminar participants at Hitotsubashi University, the University of Tokyo, Seoul National University, and the Japanese Economic Association. † Professor at the University of Tokyo. Postal Address: Hongo 7-3-1, Bunkyo-ku, Tokyo 113-0033, Japan. Email: [email protected]. Phone: +81-3-5841-5508. ‡ Professor at Seoul National University. Postal Address: Department of Economics, Seoul National University, 1 Gwanak-ro, Gwanak-gu, Seoul 08826, Korea. Email: [email protected]. Phone: +82-2-8802293. Fax: +82-2-886-4231, § Assistant Professor at Hitotsubashi University. Postal Address: 2-1, Naka, Kunitachi, Tokyo, Japan, 186-8601. Email: [email protected]. Tel: +81-42-580-8598. Fax: +81-42-580-8598.

1

1

Introduction

The causal effect of alcohol consumption on economic outcomes is of a great interest to policy-makers, particularly in terms of how alcohol consumption affects such important outcomes as health, cognitive skills, and earnings. Identifying the economic impacts of alcohol consumption is crucial to establishing related regulations and policies, such as taxes, age restrictions, and alcohol-licensing laws. In addition, the causal effect of alcohol consumption on earnings is relevant to individuals who consume alcohol. Most people believe that there are good and bad aspects of drinking: Some may worry about its adverse impact on physical health, while others regard it as helpful for social networking and fostering rapport with others. Drinking alcohol may also affect work performance positively or negatively, depending on how it is done, and indirectly via the accumulation of human capital. Given these considerations, knowing the effects of alcohol consumption on earnings and human capital is very important, not only to individuals who want to decide how best to drink, but also to labor economists. The causal impact of alcohol consumption is difficult to estimate, however, because of its endogeneity, involving such factors as reverse causality and omitted variable bias. For example, with regard to reverse causality, people with higher wages can afford more alcohol and can drink more. Concerning omitted variable bias, for example, introverted or extroverted personality characteristics may affect both alcohol consumption and wages. To date, numerous studies in economics and medical science have attempted to estimate the causal effect of alcohol consumption on wages or earnings, overcoming the endogeneity issue. Findings on the effect of alcohol on wages reported in the literature vary substantially. Some researchers observe a positive relationship between drinking and earnings (Berger and Leigh, 1988; Zarkin et al., 1998; van Ours, 2004; Tekin, 2004; Soydemir and Bastida, 2006), while others find a negative relationship between drinking and earnings (Mullahy and Sindelar, 1993; Jones and Richmond, 2006). Yet others, such as French and Zarkin (1995), Auld (2005), and Bray (2005), find an inverted U-shaped relationship, suggesting 2

the optimal amount of alcohol consumption. In addition, there have been studies finding no effect of drinking on earnings (Peters, 2004; Renna, 2008). The mixed results in the literature are partly due to differences in the estimation methods and the assumptions each estimation method relies on. Mullahy and Sindelar (1993); French and Zarkin (1995); Soydemir and Bastida (2006) applied the ordinary least squares (OLS) estimation and Jones and Richmond (2006) applied a propensity score matching method; these methods rely on the uncorrelatedness of drinking behavior and the unobserved wage determinant conditional on observed characteristics. Peters (2004) and Tekin (2004) applied a fixed effect estimation; this method relies on the uncorrelatedness of the temporal variation of drinking behavior and the temporal variation of unobserved wage determinants, conditional on the temporal variation in the observed characteristics. These assumptions are hard to justify, because the exogenous determinants of drinking behavior in level or in change are not clearly discussed. The instrumental variable method employed by various studies is more transparent in terms of the exogenous source of variation in drinking behavior. Berger and Leigh (1988) use job repetitiveness, obesity indicators, and region; Zarkin et al. (1998) use respondents’ assessments of risk variables; van Ours (2004) use early drinking and smoking indicators, the presence of a partner and children, and socioeconomic status indicators; Bray (2005) use family background at age 14 and religion; Auld (2005) use religious status and alcohol and tobacco prices; Renna (2008) used the earliest age at which drinking frequency equaled or exceeded once a month and an indicator for a family member who was a problem drinker. The validity of the instrumental variable method depends on whether the instrumental variable satisfies the exclusion restriction.1 As a credible source of exogenous variation of drinking behavior, we use genetically determined alcohol digestive ability, represented by a heterogeneous phenotype that is especially heterogeneous among East Asians. When an individual drinks alcoholic beverages, the stom1 Given that the causal relationship is estimated, studies discuss possible channels. van Ours (2004) discusses the channel of health status, while (van Ours, 2004; Bray, 2005; Tekin, 2004) point to an effect through social networking. In addition, the effect is related to schooling or labor-market productivity (Tekin, 2004; Bray, 2005; Renna, 2008).

3

ach and small intestine absorb ethyl alcohol and the liver decomposes it into acetaldehyde by alcohol dehydration (ADH) enzymes. The liver further decomposes the acetaldehyde into acetic acid by acetaldehyde dehydrogenase (ALDH) enzymes. After further decomposition into water and carbon dioxide, acetic acid is discharged to outside the body. Because of this metabolism chain, individuals with super-active ADH and/or inactive ALDH experience high acetaldehyde concentration in their blood after alcohol intake. Since acetaldehyde is highly toxic, it leads to a drunken sickness, including headache, vomiting, and a hangover. Among East Asians, because of the heterogeneity in inherited gene polymorphisms in ADH and ALDH, their alcohol digestive abilities range widely. In the East Asian population, around 40% of people are of the low active type with a weak acetaldehyde metabolism, and around 5% are of the inactive type without an acetaldehyde metabolism ability (Eng et al., 2007). The medical literature shows that people with low alcohol digestive ability are less likely to suffer from alcoholism, because they tend to drink less alcohol, as summarized by Dasgupta (2017). We use the heterogeneity of the genotype that determines alcohol digestive ability as an exogenous source of variation in alcohol drinking behaviors. We conduct original surveys of prime-aged working men in Japan and South Korea and collect a variety of information, including participants’ phenotype regarding to alcohol digestive ability, measured by a bio-marker, along with alcohol consumption (e.g., average weekly consumption, drinking frequency, beverage type) and standard socioeconomic characteristics (e.g., education, earnings, occupation). The analysis of the survey data indicates that alcohol-tolerant men drink significantly more, as well as significantly more frequently. This is consistent with findings in the existing medical literature that alcohol-tolerant people are more likely to be alcoholics, because they tend to drink more (Dasgupta, 2017). The analysis further suggests that alcohol-tolerant men receive higher hourly wages in both countries. To the best of our knowledge, our study is the first to examine the effect of alcohol digestive ability on wages. We do not, however, interpret this as evidence for the causal impact of alcohol

4

intake on hourly wage rates, because the exclusion restriction that is required to obtain an interpretable estimand through the instrumental variable method is plausibly violated. We report suggestive evidence for the violation of the exclusion restriction that alcohol-tolerant men perceive the positive impact of alcohol intake on mental health in both countries.

2

Background

2.1

Metabolism of alcohol

This study uses the interpersonal variation of alcohol digestive ability as a source of variation in drinking behaviors. To give some background information on why some East Asians are incapable of digesting alcohol, we briefly introduce how alcohol is metabolized within our body. When a person drinks alcoholic beverages, the stomach and small intestine absorb ethyl alcohol and the liver decomposes it into acetaldehyde by alcohol dehydration (ADH) enzymes. The critical gene that encodes the activity level of ADH enzyme is ADH1B; ADH1B is divided into inherited gene polymorphisms, whose types give rise to variations in the conversion efficiency of alcohol into acetaldehyde. While the polymorphism ADH1B*1 is the normal type, the polymorphism ADH1B*2 type leads to super active ADH enzymes that result in a quick acetaldehyde buildup. In the Japanese and Korean population, more than 90% of people have the ADH1B*2 gene (Eng et al., 2007). The acetaldehyde is further converted into acetic acid by acetaldehyde dehydrogenase (ALDH). Among the types of ALDH, ALDH2 plays the most important role in oxidation of acetaldehyde, and ALDH2 is divided into inherited gene polymorphisms. The polymorphism ALDH2*1 is the normal type, while ALDH2*2 encodes an inactive ALDH2 enzyme; The ALDH2 enzyme is less active among homozygotes ALDH2*2/*2 than among heterozygotes ALDH2*1/*2. Around 45% of people have ALDH2*2 in Japan and around 30% in Korea (Eng et al., 2007). Since acetaldehyde is a strongly toxic substance, it leads to a flushed face and other physical symptoms, including headache, vomiting, and a hangover. Luczak et al. (2011) 5

examine the interactive effects of ADH1B*2 and ALDH2*2 on the self-reported sensitivity to alcohol intake using Asian American college students of Chinese and Korean descent and show that ALDH2*2 (inactive acetaldehyde dehydrogenase) plays a dominant role in the determination of sensitivity to alcohol intake; they further report that given the presence of ALDH2*2, the presence of ADH1B*2 (super active alcohol dehydration) amplifies the reaction. Overall, according to existing medical literature, 30-45 % of Japanese and Koreans are sensitive to alcohol intake because of a high concentration of acetaldehyde in the bloodstream after drinking; this sensitivity to alcohol intake is specific to east Asians, including Japanese and Koreans. It is important to note that people with the low active or inactive type of ALDH virtually do not exist in European and African races; thus the study that exploits the variation of ALDH types as a source of variation in alcohol intake is best suited to the East Asian population.

2.2

Drinking together and working together

To understand the impacts of drinking on labor-market outcomes, such as earnings and hourly wages, in Japan and Korea, we should understand corporate culture and business relationships and their relationships with the activity of drinking in the two countries. To provide this background, this subsection briefly introduces the roles of alcohol drinking in business by quoting several sentences from an article written for American business managers who need to deal with Japanese or Korean business partners. Schweitzer and Kerr (2000) introduces the role of drinking alcohol in business negotiations in Japan and Korea. They first emphasize the importance of establishing a personal relationship prior to forming a formal business relationship in both countries by comparing it with Western counterparts. Easterners often strive for successful business outcomes after personal relationships have been established, while Westerners develop social relationships after business interests have been addressed. 6

As a way to establish personal relationships, business partners or colleagues at a workplace go out for drinks together. They describe the role of alcohol drinking in business transaction in Japan as follows: For centuries, Japanese business dealings of all kinds have been accompanied by drinking parties where drinking is viewed more as a ritual duty than a social pleasure. These drinking sessions can occur in large groups, as they often do in China, or in groups as small as two people. In Japan, important business meetings are often held after hours with the expectation that participants will become extremely intoxicated. In fact, many Japanese managers believe it is impossible to truly know someone without drinking heavily with them, and may feel uncomfortable with anyone who refuses to drink at a party or celebration. The following is a rather humorous description about Korea: Koreans have been known to take an especially aggressive approach to social drinking, and sometimes insist that reluctant guests take part in drinking and singing sessions. Refusing to drink without an obvious excuse may be considered rude and insulting. These gestures are taken a step further in drinking games and off-key singing performances. The more off-key the song, the greater the sense of openness and trust among participants. To refuse to drink and sing is to remain guarded and apart. That is, drinking together plays not only an important role in bridging business partners, but also an indispensable role in facilitating communications between superiors and subordinates in the workplace, as described below: The cross-level communication effects of alcohol are perhaps most clearly seen in Japanese organizations. Hierarchical relationships and protocols are especially formal, except during tsukiai, when superiors are free to give candid feedback 7

to their subordinates. These are lengthy, after-hours events typically involving large quantities of scotch. The obvious inebriation of the boss (whether actual or feigned) permits him to discuss a subordinate’s performance and shortcomings without the painful loss of face such direct feedback would entail under sober conditions. Since the exchange occurred under the influence of alcohol, both parties can come to work the next day free of the embarrassment such candor would normally produce in a Japanese organization. A subordinate, such as a teetotaling American manager who forgoes tsukiai sessions, fails to receive such performance feedback and misses an important opportunity to establish a more complete relationship with his superiors. We find a similar description of the role of drinking in business guide books for Japan and Korea (Alston and Takei, 2005; Choi and Wright, 1994). While the custom is gradually eroding and converging toward the Western style, particularly among younger generations, the drinking culture in business persists. We thus need to interpret our results in the context of this backdrop.

3

Data

It is necessary to carry out genetic testing to know the exact ALDH gene polymorphism. Fortunately, low active and inactive types can be identified through a simple and inexpensive bio-marker test, called the alcohol patch test. According to the alcohol patch test protocol, a bandage-like patch soaked with ethanol to the inner side of the upper arm for 20-30 minutes. Then, the patch determines the alcohol-digestive-ability types of individuals based on changes in skin color in the area where the patch has been removed. In this study, the alcohol patch test was distributed to our survey respondents, and the results were reported through the online survey, together with information on the frequency and per-episode amount of drinking, attitudes toward drinking, employment status, wages, working hours, and basic

8

demographic characteristics. To collect the data in Japan, we hired Intage, a web-marketing company. The target respondents were men aged 25 to 59 years who were living in Japan in March 2015. We employed stratified random sampling by age, educational background, annual income, region, industry, and employment type from the nationwide pre-registered monitors of about 210,000 people, so that the sample represents 25- to 59-year-old men in Japan. Next, we mailed the alcohol patch test along with the identification number and instructions from the web-marketing company to the selected respondents. We then conducted an online survey on the same respondents and asked them to provide the results of the alcohol patch test and answer the questions (45 items) on the various variables regarding labor-market outcomes. By putting the identification number together with the alcohol patch test, we assured that respondents had implemented the test and let them answer the questions with the results at hand. The final sample size was 2,068. The details of the alcohol patch test are as follow. First, we let respondents implement the test themselves and asked for the results of the web survey after the test. In the survey, we asked the question, “Which of the following best describes the color of the skin in the area where the patch has been removed?” For that question, each respondent selected one of the following answers: (i) the area where the patch has been removed clearly became red; (ii) the area where the patch has been removed became slightly red; (iii) the change is too little to tell whether any change actually occurred in the area where the patch has been removed, and (iv) no change in the color of the skin occurred. Herein, we categorize these four types into two; Types (i) to (iii) are categorized as the intolerant type, and Type (iv) is regarded as the tolerant type. We use the dummy variable as a proxy variable for the gene types regarding alcohol digestive ability in our analysis. We conducted a parallel survey in Korea using CNR Korea and collected responses from 518 subjects in February 2017. Figure 1 draws the distribution of the gene types measured by the results of the patch

9

test. Restricting the sample to men who had worked in the last week, and who also reported monthly earnings and the results of the alcohol patch test, the analysis sample is reduced to 1,846 from 2,068. Among 1,846 valid respondents, about 52% exhibited no change in their skin color in the alcohol patch test and are classified as the alcohol-tolerant type, about 39% are classified as sensitive, and the remaining 10% are classified as intolerant. This distribution roughly matches up with the fraction of Japanese with the ALDH2*2 gene type reported in Eng et al. (2007). Figure 2 illustrates the gene type distribution from the Korean survey. We restrict the analysis sample to men who had worked in the last week, and who also reported monthly earnings and reported the results of the alcohol patch test. Among 462 valid respondents, 60%are classified as the tolerant type, 34% as sensitive, and 5% as the intolerant type. The slightly lower fractions of respective sensitive and intolerant types than Japanese is consistent with the lower fraction of people with the ALDH2*2 gene type reported in Eng et al. (2007). In sum, both Japanese and Korean surveys find that a large fraction of subjects are alcohol sensitive or intolerant, and the distribution of the types identified from the biomarker roughly matches up with the distribution of genotypes identified by more accurate genome sequencing analysis conducted by medical scientists (Eng et al., 2007). In a further analysis, we reclassify the alcohol-tolerant type as the strong type and the alcoholsensitive and intolerant types as the weak type, as the reaction to alcohol intake is similar between homozygotes ALDH2*2/*2 and heterozygotes ALDH2*1/*2 types (Luczak et al., 2011). Table 1 tabulates the fraction of the strong type by the father’s and mother’s alcohol tolerance reported by respondents retrospectively. Among Japanese men who responded that both father and mother were the weak type, the fraction of the strong type is 25%, while the fraction is 77% if they responded that both the father and the mother were the strong type. We find a similar correlation between the parental types and one’s own type among Korean men. Chi-square tests for independence reject the null hypothesis of independence of one’s

10

own type from the father’s type in Japan and Korea; it rejects the null of the independence of the own type and mother’s type in Japan. This result confirms that the types are genetically transmitted, as indicated in the medical literature (McClearn and Kakihana, 1981; Martin et al., 1985; Heath and Martin, 1992; Bierut et al., 1998; Heath et al., 1999; Prescott and Kendler, 1999; Schuckit et al., 2001; Kendler et al., 2003; Mayfield et al., 2008). We now report the differences of drinking behaviors by the sensitivity to alcohol identified by the alcohol patch test. Figure 3 reports the average incidence or amount of alcohol drinking by the alcohol sensitivity in the Japanese sample. Strong-type men drink on about 3.6 days per week in a typical week, while weak-type men drink on about 1.8 days. When they drink, strong-type men consume about 76 ml of pure alcohol per day (in the case of beer, whose alcohol content is 5%, the number corresponds to the total volume of alcohol drinks of 1.52 liter (=76ml*100/5%*(1/1000)l)) at each occasion, while weak-type men intake 43 ml. About 29% of strong type-men are binge drinkers, who drink five glasses or more within two hours, while about 10% of the weak-type men fall into this category2 . About 16% of the strong-type men are abstainers, who do not drink at all in a typical week, while about 41% of the weak type men fall into this category. Overall, the alcohol-tolerance type identified from the alcohol patch test is a strong predictor of drinking behaviors; strong-type men drink significantly more than weak-type men in Japan. Figure 4 reports the drinking behaviors of Korean men by their alcohol-tolerance type. Similar to Japanese men, strong-type men drink more frequently than weak-type men; when strong-type men drink, they drink a larger quantity than weak-type men. Strong-type men are more likely to engage in binge drinking and are less likely to abstain from drinking than weak-type men. Compared with Japanese men, the differences of drinking behaviors between the weak and strong types are less significant among Korean men. The difference in the quantity of drinking across the two countries could be partly because of the difference in measurement; The Japanese survey asked about the consumption of all 2

The National Institute on Alcohol Abuse and Alcoholism uses this definition of binge drinking for males.

11

kinds of alcohol beverages, whereas the Korean survey asked about the consumption of beer and soju (Korean spirits). The other substantial difference between two countries is that drinking alcohol with colleagues or business partners is considered to be more important in Korea than in Japan, regardless of their own types, as evidenced in Table 3; 80% of Korean men thought that alcohol drinking facilitates communication with colleagues and supervisors, while 52% of Japanese men thought so, 73% of Korean men thought that alcohol drinking is a way to establish a good relationship with business partners, while 17% of Japanese men thought so. Thus Korean business culture evidently appreciates drinking as a way to form social bonds more than Japanese culture does. In such a business culture, Korean businessmen are presumably under stronger pressure than Japanese businessmen to drink, irrespective of their own digestive ability. Table 2 reports descriptive statistics for the samples actually used in all the regressions, dropping observations with missing variables. We do not see any obvious difference in background characteristics between the alcohol-tolerance types, except for marital status and subjective risk attitude. Weak-type men are slightly more likely to be married than strong-type men in both countries. We control for marital status in the following analysis. Regarding self-reported risk-taking attitude, the difference is not systematic; strong-type men report slightly higher risk tolerance than weak-type men among Japanese, but the relationship is reversed among Koreans. Overall, the similarity of background variables across types assures that alcohol-tolerance type is not systematically correlated with background variables.

4 4.1

Genes, Drinking, and Wages Emprical model

In this section, we estimate the effect of alcohol-tolerance type on drinking behaviors and labor-market outcomes, controlling for such basic demographic characteristics as age, edu12

cational background, and marital status. Specifically, we estimate the following models:

drinki = α toleranti + xi δ + ui

(1)

labori = β toleranti + xi ζ + vi ,

(2)

and

where drinki are the measures of drinking behaviors of individual i, such as the frequency and the amount of drinking; labori are the labor-market outcomes, such as hourly wage, hours worked in the previous week, and monthly earnings; toleranti is a variable representing alcohol tolerance; and xi includes control variables (a constant, age, its squared, indicator variables for college graduation, never married, and being married). The causal effects of alcohol tolerance on drinking behavior and labor-market outcomes are unbiased and consistently estimated when the conditional mean assumptions, E(ui |toleranti , xi ) = 0 and E(vi |toleranti , xi ) = 0, hold. These assumptions hold if alcohol tolerance is not systematically correlated with unobserved determinants of drinking behaviors or labor-market outcomes, conditional on xi (i.e. age, education, and marital status). The nonsystematic difference in the background characteristics of respondents between weak and strong types reported in Table 2 suggests that these assumptions are likely to hold. One potential identification threat is that the ALDH phenotype is inheritable, so an individual’s phenotype correlates with his or her parents’ unobserved characteristics. To overcome this potential concern, we also estimate the model that includes parents’ drinking patterns as one of the control variables. Concerning the measurement of tolerance, as explained in the data section, alcohol tolerance type is categorized as the strong type, while alcohol sensitive, intolerant, and ‘hardto-judge’ types are categorized as the weak type. As measurements of drinking behaviors, we attempt to capture the frequency of drinking and the amount of drinking. As measurements of the frequency of drinking, we use the

13

numbers of days on which respondents drank during the week prior to the survey and in a typical week. As measurement of quantity of drinking per occasion, we asked about the typical amount of drinking by each type of alcohol beverage, such as beer, sake, shochu, wine, whiskey, and other varieties in Japan. The question was limited to beer and soju in Korea. These quantities are aggregated into the pure ethanol amount, using typical alcohol contents of each variety. As a way to capture the amount of drinking in a limited time, we use whether a respondent drinks five glasses or more within two hours and refer to those who do so as binge drinkers.3 Finally, as an extreme measure of abstaining from drinking, we create a dummy variable indicating the subjects who do not drink even one day in a typical week. As measurements of labor-market outcomes, we use hourly wage, hours worked, and earnings as outcome variables. Renna (2008) hypothesized a drinking effect on wage through a reduction of working hours, because the alcohol use problem might lead to becoming sick and thus might result in fewer working hours and working days. Using the National Longitudinal Survey of the United States, she found evidence consistent with this hypothesis. To examine if the similar mechanism works in Japan and Korea, we use hourly wage, hours worked per week, and monthly earnings, all in natural logarithms, as the dependent variables labor-market outcomes.

4.2 4.2.1

Results Alcohol tolerance and drinking behaviors

We first examine the effect of respondentsf alcohol tolerance, measured by the alcohol patch test, on their drinking behaviors. Table 4 tabulates the regression results of alcohol drinking behaviors on the dummy indicating the strong type. We also used the 11-step indicator 3

Many studies have found negative impacts of binge drinking on health. For example, Welch (2017) points out that binge drinking can induce potentially severe impairments in memory and executive function, and Okoro et al. (2004) reveals that it can also induce mental distress, including stress reactions, depression, and emotional problems. Wechsler et al. (1994) find that binge drinkers are likely to cause various behavioral problems.

14

variables for alcohol tolerance to check the robustness of the results. Regardless of the choices of the dependent variable and the independent variable, results from both countries show that alcohol-tolerant men drink significantly more. Panel A of Table 4 reports the results for Japanese men. In terms of frequency of drinking, Column (3) shows that the strong type drinks about 1.8 days more than the weak type in a typical week, while the average days of drinking among the weak type is 1.8 days (upper-left panel of Figure 3). Thus, the strong type drinks twice as often as the weak type. The strong type also drinks a larger quantity in a typical drinking day. Column (7) shows that the strong type drinks 33.3 ml more than the weak type, while the average consumption of the weak type is 42.6 ml (upper-right panel of Figure 3). Thus the strong type drinks about 78% more than the weak type per day. Using the days of drinking in the last week, binge drinking behavior, or being an abstainer as the dependent variable does not change the results. Furthermore, using the 11-step measure of skill color change in the alcohol patch test qualitatively renders the same results. All the estimated differences are statistically significant at the 1 % significance level. Panel B of Table 4 reports the results for Korean men. The Korean results are qualitatively same as the Japanese results: The strong type drinks more frequently and drinks a larger amount per day. The sizes of the estimated coefficients, however, are systematically smaller than those for Japan. In term of the frequency of drinking, the difference between the strong and weak types is about 1/5 to 1/4. The difference between the two countries reconfirms the findings in Figure 3 and 4. We speculate that the difference in the results is partly due to the difference in the importance of drinking as a measure of business communication, as we discussed previously. As suggested by the subjective attitudes reported in Table 3, Korean men tend to agree with the statements that drinking alcohol facilitates social networking, communication with colleagues/supervisors, and relationships with business partners, even among the weak type. Taking the responses at face value, Korean business men are presumably under stronger pressure to drink with colleagues or business partners

15

regardless of the alcohol-tolerance type. Overall, for both countries, we robustly find that alcohol-tolerant-type men drink more in terms of frequency and amount. This finding is consistent with the received wisdom that those men with ALDH2*2, and thus those who are intolerant of alcohol consumption, drink less and consequently are protected from alcohol abuse (Dasgupta, 2017).

4.2.2

Alcohol tolerance and labor-market outcomes

We now discuss the regression results of labor-market outcomes on the alcohol-tolerance types tabulated in Table 5. Japanese results reported in Panel A show that strong-type men earn about 5.7% more than the weak type, work 3.3% less hours per week, and earn about the same amount per month. The finding that the strong type works less is consistent with the finding by Renna (2008) that drinkers work fewer hours, but the strong type earns almost the same amount per month, and consequently, the hourly wage rate is higher among strong-type men. These findings are confirmed even when we use the 11-step color change measure of the alcohol patch test. Korean results reported in Panel B are quite similar to the Japanese results in terms of the sign and the size of the estimated coefficients. Reflecting the fact that the sample size of Korea is about one quarter of Japan’s, the standard errors of the estimates become almost doubled, as predicted from the formula of the standard errors of the OLS estimators. Thus, although it is still speculative, we would argue that we would obtain the similar results to those of Japan had the Korean sample size been the same as that of Japan. It is notable that the effect of being the strong type on hourly wage is about 8.8%, which is larger than the Japanese estimate of 5.7%, given that the impact of being strong on hours worked is the same, but the impact on the monthly earnings is larger for Korea. Overall, we find evidence that being alcohol tolerant has a positive impact on hourly wage in Japan, and we also find suggestive evidence for that in Korea too. Given that the difference of drinking behaviors between the strong and weak types is significantly smaller,

16

the Wald estimator formula suggests that the effect of drinking on hourly wage is larger in Korea than in Japan. While this argument is not totally out of the ballpark, in the next section, we discuss that we cannot exactly estimate a meaningful estimand of the causal impact of drinking on hourly wage.

5

Why not use an instrumental variable estimation?

The empirical strategy in this research is very closely related to the concept of “Mendelian randomization,” which involves a method of using variations in the genes of known function to examine the causal effects of modifiable exposure on disease, as advocated by epidemiologists (Smith and Ebrahim, 2003, 2004, 2005; Smith, 2010). As an example of a study published in an economics journal, Zuccolo et al. (2013) has studied the association between mothers’ drinking and offspring school outcomes assuming that (1): the ADH1B genotype (resulting in efficient transformation of alcohol into acetaldehyde) is associated with maternal alcohol consumption, (2): there is no association between the genotype and confounding factors, and (3): the genotype does not affect outcomes through any path other than maternal alcohol consumption. By testing the maternal ADH1B-offspring outcome association, Zuccolo et al. (2013) examined the causal impact of maternal alcohol use on offspring outcomes. Their analysis shows that children whose mothers have a lower alcohol consumption genotype have higher outcomes than children whose mothers have heavier drinking genes. This implies that drinking during pregnancy has a potential negative impact on childrenfs educational outcomes. Given the idea of Mendelian randomization and our previous estimation results, it is tempting to estimate the causal effect of drinking on labor-market outcomes using the alcohol tolerance type as the instrumental variable. In a constant treatment effect setting, if the genetic polymorphism of ALDH meets the excluded variable restriction, i.e., the patch results affect only the outcome variables through drinking behavior, it should be possible to perform a consistent estimation for drinking’s causal influence on labor-market outcomes by 17

the instrument variable estimation. The similarity of background characteristics between the alcohol tolerant and intolerant groups suggests that the exclusion restriction is satisfied and that the causal impact is estimable via the instrumental variable estimation. The estimation of a meaningful causal impact, however, becomes impossible under a more realistic assumption that the causal impact of alcohol consumption on labor-market outcomes depends on the alcohol tolerance type and thus the treatment effects are heterogeneous and dependent on the gene types. We now examine the case when the treatment effects of drinking on labor-market outcomes are heterogeneous across individuals and the individualfs specific treatment effect depends on the alcohol tolerance type. Let βi be an individualfs specific treatment effect of alcohol consumption on labor-market outcomes and β¯ be the average treatment effect over the population. The model can be written as:

labori = βi drinki + xi θ + ei ¯ drinki ], = β¯ drinki + xi θ + [ei + (βi − β)

(3) (4)

and drinki = α toleranti + xi δ + ui .

(5)

This is a situation to estimate the average treatment effect in the random coefficient model by the two stage least squares that has been considered in the literature (Heckman and Vytlacil, 1998; Wooldridge, 1997, 2003). The instrumental variable estimator is a consistent estimator of β¯ if a) the first stage is relevant: α 6= 0 and b) the exclusion restriction is satis¯ drinki ]} = 0. We confirmed that alcohol tolerance affects the fied: E{toleranti [ei + (βi − β) frequency and amount of drinking in Table 4, and thus the first-stage assumption is satisfied. The exclusion restriction is a controversial assumption in our context. Assuming that the alcohol tolerance affects labor-market outcomes only through drinking behavior is a reason-

18

¯ drinki ]} = 0, however, able assumption, thus E[toleranti ei ] = 0 holds. E{toleranti [(βi − β) may not hold. A sufficient condition for this condition to hold is Cov(βi , drinki |toleranti ) = Cov(βi , drinki ), as articulated by Wooldridge (2003).4 In a likely scenario that the association between the return to drinking and the frequency/amount of drinking is stronger among the strong type than the weak type, the assumption is violated and the exclusion restriction is violated. In the end, given α > 0, the plausible violation of the exclusion restriction ¯ drinki ]} > 0 results in an asymptotic upward bias in a form of E{toleranti [ei + (βi − β) ¯ In other words, if the association of the instrumental variable estimator, i.e., plim βˆ¯ > β. of the labor market return to drinking and drinking frequency/amount is stronger among alcohol-tolerant people, the instrumental variable estimation asymptotically overestimates the average treatment effect. We now provide suggestive evidence that the labor-market return is plausibly higher among the strong type than the weak type. Our survey asked for respondents’ opinions on the benefits and costs of alcohol consumption. Table 3 tabulates the responses by the alcohol tolerance types. In both countries, about 40% of men think that drinking alcohol is good for mental health, and those of the strong type are 12-13 percentage points more likely to support this opinion. In Japan, strong-type men are more likely to agree with the statements in statistically significant ways than the weak type men that alcohol consumption relieves stress, facilitates communication with colleagues or supervisors, and has a good effect on physical health. Similarly in Korea, strong-type men are more likely to agree with the statements that alcohol consumption enhances social networking, facilitates communication with colleagues or supervisors, and strengthens relationships with business partners. While strong-type Japanese men are more likely to agree with the statement that drinking wastes time, overall, strong-type men in both countries are more likely than weak-type men to agree with the statements indicating the positive effects of alcohol consumption on several outcomes. This 4 ¯ drinki ]} = E{toleranti E[(βi − By the law of interated expectations, E{toleranti [(βi − β) ¯ drinki |toleranti ]} = E[toleranti Cov(βi , drinki |toleranti )] = E(toleranti )Cov(βi , drinki ). Since the last β) expression is a constant, adjusting for the constant term of the estimating equation assures the consistency ˆ ¯ of β.

19

finding suggests that the return to alcohol drinking on labor-market outcomes could be higher among alcohol-tolerant men and the violation of the exclusion restriction. One still might think that the instrumental variable estimator consistently estimates the local average treatment effects among compliers, but again, it is impossible because of the plausible violation of the exclusion restriction. We briefly show why this is the case in Rubin’s counterfactual framework. Let zi be the indicator for the strong type, di be the indicator for drinkers, and yi be labor-market outcomes. Angrist et al. (1996) show that under the assumptions of a) the relevance of the instrumental variable: E(di |zi = 1) > E(di |zi = 0), b) the exclusion restriction yi (di , zi = 0) = yi (di , zi = 1) = yi (di ) for di = {0, 1}, and c) monotonicity di (zi = 1) ≥ di (zi = 0) for all i, the instrumental variable estimator is a consistent estimator of the local average treatment effect among compliers, E(yi (di = 1) − yi (di = 0) | i ∈ {di (zi = 1) = 1 and di (zi = 0) = 0}). The previous results show that the instrumental variable is relevant and the monotonicity assumption may well hold. Thus whether the instrumental variable consistently estimates the local average treatment effect depends on the validity of the exclusion restriction. Assuming the invariance of the labor-market outcome when an individual does not drink regardless of alcohol tolerance type, i.e., yi (di = 0, zi = 0) = yi (di = 0, zi = 1) is not controversial, but the assumption when an individual drinks, i.e. yi (di = 1, zi = 0) = yi (di = 1, zi = 1) would be controversial. This assumes that the labor-market outcome of individual i does not depend on the alcohol tolerance type when he drinks. The survey questions regarding the effect of alcohol consumption arguably capture the subjective feeling of yi (di = 1, zi ) − yi (di = 0, zi ), given own type zi . The findings of Table 3 indicate that strong-type men feel a more positive effect from drinking than weak-type men, on average, and thus E[yi (di = 1, zi = 1) − yi (di = 0, zi = 1)] > E[yi (di = 1, zi = 0) − yi (di = 0, zi = 0)]. This result suggests the violation of the exclusion restriction. The reason why we cannot estimate the local average treatment effect can be understood by considering who the compliers are in our context. Now compliers include those who drink

20

because they can di (zi = 1) = 1 and those who do not drink because they cannot di (zi = 0) = 0. These two groups could be heterogeneous in terms of the treatment effects, and defining a single estimand of the treatment effects over these two groups is apparently difficult. In other words, medical doctors would have hard time giving general advice regarding drinking alcohol to these two heterogeneous populations. Epidemiologists have already recognized the limitation of the Menderian randomization. Lewis and Smith (2005) surveyed the literature on the relationship between ALDH2 genotype and the episode of esophageal cancer and reported that the risk of esophageal cancer is reduced among ALDH2*2*2 homozygotes but increased among ALDH2*1*2 heterozygotes, relative to ALDH2*1*1 homozygotes. They have pointed out that the genotype of ALDH2 influences esophageal cancer risk through two mechanisms: first through influencing alcohol intake and second through influencing acetaldehyde levels. While they do not articulate it in their paper, the second channel implies the dependence of the treatment effects of alcohol consumption on esophageal cancer risk on the ALDH2 genotype; the negative impact of alcohol intake on the esophageal cancer risk is more significant among ALDH2*2 types and thus violates the exclusion restriction. Lewis and Smith (2005) recognized this as a limitation of the Menderian randomization approach.

6

Why do those who can drink earn more?

While we cannot establish a causal relationship of drinking on wages, we know that those who can drink earn more. In this section, we attempt to understand the mechanism behind the causal relationship of alcohol tolerance on labor-market outcomes. To attain this goal, we estimate the alcohol effect on various outcomes. There should be various channels through which alcohol consumption affects wages and hours, such as effects on both mental and physical health and networking effects obtained through drinking with people. Table 7 reports the differences of the proxy variables for physical and mental health and social networking. As a health measure, we first use the 5-point Likert scale of health 21

condition (5: Good - 1: Bad) as the measurement. Table 7 reports the mean differences of health condition between the strong and weak types. While the strong type in Japan is healthier than the weak type in Japan, we do not find a systematic difference in Korea. Next, we examine the difference of Body Mass Index (BMI) and find that the strong type is slightly but statistically significantly heavier than the weak type. As a way to examine the difference in mental health status, we also compare the hours of sleep, self-reported happiness, and life satisfaction by the alcohol tolerance types. For the hours of sleep, we find no systematic difference, but among Koreans, we find that the strong-type men are significantly happier and more satisfied with their lives than the weak type. While the ways it shows up in the responses to the survey question differ across countries, overall, we find suggestive evidence that the strong type is healthier than the weak type, both physically and mentally in each country. Since a significant number of respondents agreed that drinking facilitates communication with colleagues or business partners, as reported in Table 3, men with alcohol tolerance may select into occupations that require working with people or establishing a strong network with surrounding people. To capture the number of people with whom the respondents work on a daily basis, we use the responses to the following question “at your everyday work, how many people (coworkers, whether rank is higher or lower) do you work together with?” Table 7 reports that there are no systematic differences in the responses to this question between strong and weak types in Japan. In contrast, in Korea, strong-type men work with 8 more people, whereas weak-type men work with 12 people, on average. Another way to capture the occupational selection by the alcohol tolerance type is to use the 5-point Likert scale responses (5: strongly agree - 1: strongly disagree) to the question that asks if the respondentfs job requires interpersonal skill. For this question, we do not find a significant difference in responses between the strong and weak types. Finally, to capture the difference in the depth of the social network, we use the number of people who could lend 500,000 yen to the respondents if in need of money. Again, we do not find significant differences between

22

the two types in Japan, but we find that strong-type men have 1 person, while weak-type men have 3 people, on average, in Korea. While there are a few notable differences in the health status or the measurement of social capital, the difference between the two groups is not systematic, and we cannot pin down the common mechanism for why the alcohol-tolerant type earns more across the two countries. The results suggest, however, that different mechanisms are at work in Japan and Korea. In Japan, the benefit comes through the health channel, while in Korea the benefit comes though the social network channel. This difference is consistent with the difference found in the attitudes toward drinking between Japan and Korea in Table 3, indicating that only Koreans recognize that drinking is beneficial for social networking.

7

Conclusion

In this study, we aimed to estimate the causal impact of alcohol tolerance on drinking behavior and wages. We conducted original surveys covering about 2,000 and 500 prime-age men in Japan and Korea, respectively; the survey collected information on a bio-marker test result that identifies subjects’ tolerance, frequency and amount of alcohol consumption, hours worked, and monthly earnings, along with other demographic variables. We found that alcohol tolerance is a strong predictor of alcohol consumption, but it is not systematically correlated with background variables. Furthermore, we found that those with alcohol tolerance earn higher wages. While numerous medical studies find that those who are intolerant to alcohol are less likely to drink and as a consequence are less likely to suffer from alcoholism, as summarized in the survey paper by Dasgupta (2017), to the best of our knowledge, we are the first to find evidence that alcohol tolerant men earn more than intolerant men. Furthermore, we provided suggestive evidence that alcohol-tolerant men earn higher wages because they keep better mental health by releasing stress by drinking, and they establish better relationships with colleagues and business partners. Although it is tempting to infer the causal impact of drinking on wages from the results 23

using the instrumental variable estimation, we discussed that we cannot establish such a causal relationship because of the violation of the exclusion restriction. To understand this, we should imagine an individual who actually does not drink with a certain labor market outcome. The exclusion restriction requires that the counterfactual labor-market outcome when he drinks should be identical regardless of his alcohol tolerance. This is a strong assumption that is difficult to justify, and we provide suggestive evidence consistent with the violation of the exclusion restriction. While Mendelian randomization is a powerful tool to identify the causal impact of an environmental exposure on an outcome, to obtain an estimator allowing for a meaningful interpretation, the treatment effects should not be correlated with the genetic type that determines the environmental exposure. The difficulty in establishing causal effects of alcohol drinking on labor-market outcomes among East Asians implies the danger of policy makers giving singular advice on the proper amount of alcohol intake without paying attention to the heterogeneous reactions to drinking alcohol. Tailor-making advice according to individual genetic type is unrealistically costly, but the advice should be at least tailored based on the alcohol-tolerance type that is identified by a simple bio-marker or even more simply by observing the face flushing after a small amount of alcohol intake. Our study is limited by a small sample size and measurement error because of simple bio-marker usage. We hope that this study will be connected to a bigger project to generate a larger sample with accurate genome assay testing information to obtain more accurate results.

References Alston, J.P. and I. Takei, Japanese Business Culture and Practices: A Guide to Twentyfirst Century Japanese Business, iUniverse, 2005. Angrist, Joshua D., Guido W. Imbens, and Donald B. Rubin, “Identification of Causal Effects Using Instrumental Variables,” Journal of the American Statistical Association, 1996, 91 (434), 444–455.

24

Auld, M. Christopher, “Smoking, Drinking, and Income,” Journal of Human Resources, 2005, 40 (2), 505–518. Berger, Mark C. and J. Paul Leigh, “The Effect of Alcohol Use on Wages,” Applied Economics, 1988, 20 (10), 1343–1351. Bierut, L.J., S.H. Dinwiddie, H. Begleiter, R.R. Crowe, V. Hesselbrock, JI Jr Nurnberger, B Porjesz, M.A. Schuckit, and T. Reich, “Familial transmission of substance dependence: alcohol, marijuana, cocaine, and habitual smoking,” Archives of General Psychiatry, 1998, 55 (11), 982–988. Bray, Jeremy W., “Alcohol Use, Human Capital, and Wages,” Journal of Labor Economics, 2005, 23 (2), 279–312. Choi, C.J. and N. Wright, How to Achieve Business Success in Korea: Where Confucius Wears a Three-Piece Suit, Palgrave Macmillan UK, 1994. Dasgupta, Amitava, “6 - Genetic polymorphisms of alcohol metabolizing enzymes associated with protection from or increased risk of alcohol abuse,” in Amitava Dasgupta, ed., Alcohol, Drugs, Genes and the Clinical Laboratory, Academic Press, 2017, pp. 107 – 116. Eng, Mimy Y, Susan E Luczak, and Tamara L Wall, “ALDH2, ADH1B, and ADH1C Genotypes in Asians: A Literature Review,” Alcohol Research & Health, 2007, 30 (1), 22–27. French, Michael T. and Gary A. Zarkin, “Is Moderate Alcohol Use Related to Wages? Evidence from Four Worksites,” Journal of Health Economics, 1995, 14 (3), 319–344. Heath, A.C. and N.G. Martin, “Genetic differences in alcohol sensitivity and the inheritance of alcoholism risk,” Journal of Studies on Alcohol, 1992, 53 (2), 262–271. , P.A. Madden, K.K. Bucholz, S.H. Dinwiddie, W.S. Slutske, L.J. Bierut, J.W. Rohrbaugh, D.J. Statham, M.P. Dunne, J.B. Whitfield, and N.G. Martin, “Genetic differences in alcohol sensitivity and the inheritance of alcoholism risk,” Psychological Medicine, 1999, 29, 1069–1081. Heckman, James and Edward Vytlacil, “Instrumental Variables Methods for the Correlated Random Coefficient Model: Estimating the Average Rate of Return to Schooling When the Return is Correlated with Schooling,” The Journal of Human Resources, 1998, 33 (4), 974–987. Jones, Alison Snow and David W. Richmond, “Causal Effects of Alcoholism on Earnings: Estimates from The NLSY,” Health Economics, 2006, 15 (8), 849–871. Kendler, K.S., K.C. Jacobson, C.A. Prescott, and M.C. Neale, “Specificity of genetic and environmental risk factors for use and abuse/dependence of cannabis, cocaine, hallucinogens, sedatives, stimulants, and opiates in male twins,” American Journal of Psychiatry, 2003, 160, 687–695.

25

Lewis, Sarah J. and George Davey Smith, “Alcohol, ALDH2, and Esophageal Cancer: A Meta-analysis Which Illustrates the Potentials and Limitations of a Mendelian Randomization Approach,” Cancer Epidemiology and Prevention Biomarkers, 2005, 14 (8), 1967–1971. Luczak, Susan E., Danielle Pandika, Shoshana H. Shea, Mimy Y. Eng, Tiebing Liang, and Tamara L. Wall, “ALDH2 and ADH1B Interactions in Retrospective Reports of Low-Dose Reactions and Initial Sensitivity to Alcohol in Asian American College Students,” Alcoholism: Clinical and Experimental Research, 2011, 35 (7), 1238–1245. Martin, N.G., J.G. Oakeshott, J.B. Gibson, G.A. Starmer, J. Peri, and A.V. Wilks, “A twin study of psychomotor and physiological responses to an acute dose of alcohol,” Behavior Genetics, 1985, 15 (4), 305–347. Mayfield, R.D., R.A. Harris, and M.A. Schuckit, “Genetic factors influencing alcohol dependence,” British Journal of Pharmacology, 2008, 154 (2), 275–287. McClearn, G.E. and Ryoko Kakihana, “Selective breeding for ethanol sensitivity: Short-sleep and long-sleep mice,” in G.E. McClearn, R.A. Deitrich, and V.G. Erwin, eds., Development of animal models as pharmacogenetic tools, Washington, D.C.: DHHS Publication No. [ADM] 81-113, U.S. Government Printing Office, 1981, pp. 147–159. Mullahy, John and Jody L. Sindelar, “Alcoholism, Work, and Income,” Journal of Labor Economics, 1993, 11 (3), 494–520. Okoro, Catherine A., Robert D. Brewer, Timothy S. Naimi, David G. Moriarty, Wayne H. Giles, and Ali H. Mokdad, “Binge drinking and health-related quality of life: do popular perceptions match reality?,” American Journal of Preventive Medicine, 2004, 26 (3), 230–233. Peters, Bethany L, “Is There A Wage Bonus from Drinking? Unobserved Heterogeneity Examined,” Applied Economics, 2004, 36 (20), 2299–2315. Prescott, C.A. and K.S. Kendler, “Genetic and environmental contributions to alcohol abuse and dependence in a population-based sample of male twins,” American Journal of Psychiatry, 1999, 156, 34–40. Renna, Francesco, “Alcohol Abuse, Alcoholism, and Labor Market Outcomes: Looking for the Missing Link,” Industrial and Labor Relations Review, 2008, 62 (1), 92–103. Schuckit, M.A., H.J. Edenberg, J. Kalmijn, L. Flury, T.L. Smith, T. Reich, L. Bierut, A. Goate, and T. Foroud, “A genome-wide search for genes that relate to a low level of response to alcohol,” Alcoholism: Clinical and Experimental Research, 2001, 25, 323–329. Schweitzer, Maurice E. and Jeffrey L. Kerr, “Bargaining under the Influence: The Role of Alcohol in Negotiations,” The Academy of Management Executive (1993-2005), 2000, 14 (2), 47–57. 26

Smith, George Davey, “Mendelian Randomization for Strengthening Causal Inference in Observational Studies: Application to Gene Environment Interactions,” Perspectives on Psychological Science, 2010, 5 (5), 527–545. and Shah Ebrahim, “‘Mendelian Randomization’: Can Genetic Epidemiology Contribute to Understanding Environmental Determinants of Disease?,” International Journal of Epidemiology, 2003, 32 (1), 1–22. and , “Mendelian Randomization: Prospects, Potentials, and Limitations,” International Journal of Epidemiology, 2004, 33 (1), 30–42. and , “What can mendelian randomisation tell us about modifiable behavioural and environmental exposures?,” BMJ, 2005, 330 (7499), 1076–1079. Soydemir, Gke A. and Elena Bastida, “Alcohol Use and Earnings: Findings from a Community Based Study,” Eastern Economic Journal, 2006, 32 (4), 617–628. Tekin, Erdal, “Employment, Wages, and Alcohol Consumption in Russia,” Southern Economic Journal, 2004, 71 (2), 397–417. van Ours, Jan C., “A Pint A Day Raises A Man’s Pay; But Smoking Blows That Gain Away,” Journal of Health Economics, 2004, 23 (5), 863–886. Wechsler, Henry, Andrea Davenport, George Dowdall, Barbara Moeykens, and Sonia Castillo, “Health and behavioral consequences of binge drinking in college. A national survey of students at 140 campuses,” Journal of the American Medical Association, 1994, 272 (21), 1672–1677. Welch, Killian A, “Alcohol consumption and brain health,” British Medical Journal, 2017, 357:j2645. Wooldridge, Jeffrey M, “On two stage least squares estimation of the average treatment effect in a random coefficient model,” Economics Letters, 1997, 56 (2), 129 – 133. Wooldridge, Jeffrey M., “Further results on instrumental variables estimation of average treatment effects in the correlated random coefficient model,” Economics Letters, 2003, 79 (2), 185 – 191. Zarkin, Gary A., Michael T. French, Thomas Mroz, and Jeremy W. Bray, “Alcohol Use and Wages: New Results from The National Household Survey on Drug Abuse,” Journal of Health Economics, 1998, 17 (1), 53–68. Zuccolo, Luisa, Sarah J Lewis, George Davey Smith, Kapil Sayal, Elizabeth S Draper, Robert Fraser, Margaret Barrow, Rosa Alati, Sue Ring, John Macleod, Jean Golding, Jon Heron, and Ron Gray, “Prenatal Alcohol Exposure and Offspring Cognition and School Performance. A ‘Mendelian Randomization’ Natural Experiment,” International Journal of Epidemiology, 2013, 42 (5), 1358–1370.

27

Figure 1: Distribution of alcohol tolerance types in Japan

0

Fraction of respondents (N = 1,846) .1 .2 .3 .4

.5

0.518

0.385

0.096

Intolerant

Sensitive

Tolerant

Figure 2: Distribution of alcohol tolerance types in South Korea

Fraction of respondents (N = 462) .2 .4

.6

0.606

0.344

0

0.050

Intolerant

Sensitive

28

Tolerant

Figure 3: Sensitivity to alcohol and drinking behavior in Japan Amount of alcohol per day (ml)

4

80

Drinking days of a typical week

76.294

3

60

3.560

40 20

Weak type

0

0

1

2

42.633 1.825

Strong type

Binge (Drink 5 glasses or more in 2 hours)

0.414

.4 .3 .2

.2

0.155

0.101

Weak type

0

0

.1

.1

Strong type

Abstainer (Do not drink in a typical week)

0.285

.3

Weak type

Strong type

Weak type

Strong type

Figure 4: Sensitivity to alcohol and drinking behavior in South Korea Drinking days of a typical week

Amount of alcohol per day (ml) 80

2

1.946

66.434

0

.5 0

Weak type

Strong type

Weak type

Strong type

Abstainer (Do not drink in a typical week) .2

1

Binge (Drink 5 glasses or more in two hours) 0.900

0.163

.15

0.766

.1

.6

.8

40.467

20

1

40

60

1.5

1.609

Weak type

0

0

.2

.05

.4

0.075

Strong type

29

Weak type

Strong type

Table 1: Fraction of strong type given parental type

Mother type Weak Strong

Japan Father type Weak Strong 0.246 0.382 (460) (186) 0.534 0.771 (592) (463)

Korea Father type Weak Strong 0.468 0.500 (154) (26) 0.675 0.711 (228) (45)

216.2 < 0.01 105.8 < 0.01

19.8 < 0.01 0.655 0.418

Chi-square test for independence with father’s type p-value with mother’s type p-value

30

Table 2: Descriptive Statistics

Japan (N=1,882) Weak Strong-Weak Father’s college

0.29

Mother’s college

0.18

Age College educated

42.47 (9.36) 0.66

Married

0.63

Never married

0.32

Health (1-5) (1=Good, 5=Bad) Smoking

3.32 (0.95) 0.27

Risk (0-10) (0=Intorelant, 10=Torelant) Hours of sleep

3.85 (2.42) 6.26 (1.15) 0.86

Wage and salaried Hourly wage Work hours Monthly earnings

2.32 (1.98) 46.42 (12.44) 431.64 (243.60)

0.03 [0.02] 0.04 [0.02] -0.28 [0.43] -0.01 [0.02] -0.04 [0.02] 0.03 [0.02] 0.09 [0.04] -0.02 [0.02] 0.14 [0.11] -0.00 [0.05] 0.01 [0.02] 0.18 [0.10] -0.52 [0.60] 7.53 [11.50]

Korea (N = 464) Weak Strong-Weak 0.35 0.20 41.99 (9.12) 0.63 0.67 0.32 3.15 (0.77) 0.46 4.72 (1.85) 6.63 (0.99) 0.79 2.16 (1.50) 52.24 (16.99) 449.18 (296.63)

0.03 [0.05] -0.01 [0.04] -2.42 [0.85] 0.01 [0.05] -0.09 [0.05] 0.08 [0.05] 0.02 [0.08] -0.11 [0.05] -0.55 [0.18] -0.08 [0.09] 0.01 [0.04] 0.02 [0.15] -0.71 [1.64] -15.17 [26.85]

Notes: Standard deviations are in parenthesis and heteroskedasticity robust standard errors are in blackets below estimates. Hourly wages and monthly earnings are before-tax monthly income in 1,000 JPN or 10,000 KRW, which are close to 10USD. The Korean survey asked the monthly earnings in 8 intervals. We used mid-points and topcoded as the last interval’s lower bound + 50% extra.

31

Table 3: Attitudes toward drinking by alcohol tolerance type (1) (2) (3) Japan (N=1,882) Weak Strong Adj. -Weak Relieve stress 0.232 0.155 0.156 (0.021) (0.021) Social network 0.586 0.018 0.020 (0.023) (0.023) Communication with colleagues or supervisors 0.497 0.035 0.039 (0.023) (0.023) Relationship with business partners 0.163 0.020 0.021 (0.017) (0.017) Good effect on physical health 0.084 0.027 0.026 (0.014) (0.014) Good effect on mental health 0.328 0.133 0.132 (0.022) (0.022) Bad effect on physical health 0.262 -0.013 -0.015 (0.020) (0.020) Bad effect on mental health 0.106 -0.027 -0.028 (0.013) (0.013) Drinking wastes time 0.149 0.031 0.029 (0.017) (0.017)

(4)

(5) (6) Korea (N=464) Weak Strong Adj. -Weak 0.337 0.034 0.044 (0.045) (0.046) 0.777 0.076 0.084 (0.037) (0.038) 0.772 0.050 0.067 (0.039) (0.038) 0.701 0.049 0.073 (0.043) (0.042) 0.011 0.011 0.017 (0.025) (0.025) 0.353 0.097 0.124 (0.046) (0.045) 0.793 -0.000 -0.005 (0.039) (0.039) 0.489 -0.050 -0.057 (0.047) (0.047) 0.353 -0.057 -0.062 (0.044) (0.045)

Notes: Columns (1) and (4) present means of dummy variables. Columns (2) and (5) present the differences of means between strong and weak types with heteroskedasticity robust standard errrors in parenthesis. Columns (3) and (6) present the estimates of the coefficient for the strong type controlling for age, age squared, dummy variables indicating college graduation, being married and being never-married.

32

33 464 0.026

0.417*** (0.139)

1,882 0.160

1.778*** (0.109)

464 0.027

0.074*** (0.025)

1,882 0.192

0.299*** (0.015)

464 0.028

0.356*** (0.125)

1,882 0.160

1.757*** (0.109)

464 0.026

0.057** (0.023)

1,882 0.186

0.290*** (0.016)

(3) (4) Drinking days in a typical week

(6)

464 0.039

0.138*** (0.037)

1,882 0.064

0.185*** (0.018)

464 0.058

0.030*** (0.007)

1,882 0.065

0.028*** (0.002)

Binge drinking

(5)

(8)

464 0.059

25.832*** (5.023)

1,880 0.042

33.290*** (4.060)

464 0.060

4.465*** (0.881)

1,880 0.055

5.862*** (0.547)

Alcohol amount

(7)

(10)

464 0.032

-0.086*** (0.032)

1,882 0.091

-0.262*** (0.020)

464 0.042

-0.019*** (0.006)

1,882 0.127

-0.047*** (0.003)

Abstainer

(9)

Notes: Control variables (age, age squared, college, never married, and married) are included. Robust standard errors are presented in parentheses. *** 1%; ** 5%; * 10% significance.

Observations R-squared

Scale 0-10 (reversed)

Observations R-squared B. Korea Strong type

Scale 0-10 (reversed)

A. Japan Strong type

(1) (2) Drinking days in the last week

Table 4: ALDH Genotypes and Alcohol Consumption

Table 5: ALDH Genotypes and Labor Market Outcomes

A. Japan Strong type

(1) (2) Hourly wage (log)

(3) (4) Work hours (log)

0.057** (0.027)

-0.033* (0.018)

Scale 0-10 (reversed) Observations R-squared B. Korea Strong type

0.012*** (0.004) 1,871 0.202 0.088* (0.053)

Scale 0-10 (reversed) Observations R-squared

1,871 0.204

0.024 (0.026) -0.006** (0.003)

1,871 0.026

1,871 0.027

-0.031 (0.032) 0.014 (0.009)

464 0.251

464 0.250

(5) (6) Earnings (log)

0.006 (0.004) 1,871 0.270 0.056 (0.047)

-0.006 (0.006) 464 0.043

1,871 0.270

464 0.043

0.008 (0.008) 464 0.330

464 0.330

Notes: Control variables (age, age squared, college, never married, and married) are included. Robust standard errors are presented in parentheses. *** 1%; ** 5%; * 10% significance.

34

Table 6: ALDH Genotypes and Labor Market Outcomes among Wage and Salaried Workers

A. Japan Strong type

(1) (2) Hourly wage (log)

(3) (4) Work hours (log)

0.058** (0.027)

-0.033* (0.017)

Scale 0-10 (reversed) Observations R-squared B. Korea Strong type

0.013*** (0.004) 1,616 0.212 0.112** (0.055)

Scale 0-10 (reversed) Observations R-squared

1,616 0.215

0.025 (0.026) -0.006** (0.003)

1,616 0.027

1,616 0.028

-0.036 (0.037) 0.021** (0.010)

368 0.201

368 0.202

(5) (6) Earnings (log)

0.006* (0.004) 1,616 0.293 0.076* (0.043)

-0.005 (0.007) 368 0.047

1,616 0.294

368 0.046

0.016** (0.008) 368 0.278

368 0.281

Notes: The sample is restricted to wage and salaried workers only. Control variables (age, age squared, college, never married, and married) are included. Robust standard errors are presented in parentheses. *** 1%; ** 5%; * 10% significance.

35

Table 7: Possible Channels (1) (2) (3) Japan (N = 1,882) Weak Strong Adj -Weak Healthy (Scale:1-5) 3.319 0.091 0.093 (0.949) (0.044) (0.044) BMI 24.535 1.400 1.556 (34.910) (2.864) (2.953) Sleeping hours 6.258 -0.002 -0.005 (1.154) (0.050) (0.050) Happy (Scale:1-5) 3.377 0.033 0.066 (0.988) (0.046) (0.043) Satisfied with the current life (Scale:1-5) 3.104 0.057 0.080 (1.054) (0.048) (0.046) No. of people at work 19.925 0.039 0.224 (27.884) (1.387) (1.391) Interpersonal skills required 0.842 0.013 0.015 (0.365) (0.017) (0.016) No. of people you can borrow 0.5 million yen 2.253 0.102 0.137 (4.317) (0.212) (0.212)

(4) (5) (6) Korea (N = 464) Weak Strong Adj -Weak 3.152 0.023 0.021 (0.767) (0.078) (0.079) 24.193 0.493 0.570 (3.020) (0.285) (0.287) 6.630 -0.084 -0.099 (0.989) (0.093) (0.095) 3.353 0.157 0.148 (0.810) (0.077) (0.078) 3.255 0.191 0.179 (0.833) (0.080) (0.080) 12.272 8.560 8.416 (12.537) (4.204) (4.349) 0.783 -0.015 -0.004 (0.414) (0.040) (0.040) 2.815 1.185 1.315 (2.438) (0.478) (0.557)

Notes: Columns (1) and (4) present sample mean and standard deviation. Columns (2) and (5) present the mean difference between strong and weak types. Columns (3) and (6) present regression coefficient for strong type dummy variable controlling for age, age squared, college, never married, and married.

36